Type
(BT,MT,RI) | Topic
(with short description) | Contact | possible
start date | Time Topic Added |
|---|
| MT, RI | Digital Twin for Building Thermodynamics and Electrical Appliances Research method: Prototyping Research questions: How do different modeling approaches, including white-box (e.g., OCHRE), grey-box (e.g., RC models), and black-box (e.g., Random Forrests, LSTMs, Transformers, etc.) models, perform in optimization frameworks for smart building energy management systems (BEMS)? - What are the challenges creating digital twins of buildings from a data perspective?
- How to make existing ("dumb") buildings smart / How can they be modeled such that a model for optimization can be created?
Possible approach: | Paul Loer paul.loer@tum.de | anytime | 2026-04 |
| MT, RI | Optimal Siting of AI Data Centers Research method: Prototyping Research question: How can AI data centers be optimally sited by co-optimizing electric grid capacity, collocated renewable generation, and critical infrastructure like water and telecommunications? How can physical grid modeling identify "hidden capacity" in existing networks to significantly accelerate data center grid connections? How does the mandatory integration of data center waste heat into local thermal networks impact capacity expansion models and site selection?
Possible approach: Develop an optimization framework (e.g., in Python) that evaluates data center sites based on electric grid constraints, renewable potential, and infrastructure costs. Integrate thermo-electric co-optimization to model the physical restrictions of both the electricity grid and local heat networks simultaneously. Benchmark the developed model using distribution grid datasets to quantify the potential reduction in connection times and infrastructure requirements.
| Manuel Katholnigg manuel.katholnigg@tum.de | anytime | 2026-04 |
| MT, RI | LLM-Assisted Tariff Parsing for Local Energy Management Systems Research method: Prototyping Research question: How effectively can Large Language Models (LLMs) parse complex, multi-page industrial electricity tariff PDFs into functional Energy Management System (EMS) logic? How can this translation process be executed entirely locally to comply with legal constraints regarding cloud uploads?
Possible approach: Investigate LLM-assisted methods for parsing and extracting rules from complex tariff documents. Develop a fully local open-source pipeline that accepts a tariff PDF and generates a functional EMS configuration without data leaving the plant. Benchmark the rule extraction accuracy and performance across various types of electricity tariffs.
| Manuel Katholnigg manuel.katholnigg@tum.de | anytime | 2026-03 |
| MT, RI | Automated Discovery of Industrial Demand Response Potential Research method: Prototyping Research question: How can industrial demand response potential be automatically quantified directly from load curves What is the expected economic value of this flexibility across different electricity markets (e.g., Intraday, Balancing, DSO)?
Possible approach: Utilize clustering and anomaly detection algorithms to analyze industrial 15-minute load curve CSV data. Develop an economic benchmark to evaluate the cross-sector and cross-region flexibility potential. Build a streamlined automated pipeline that calculates the expected annual revenue based on the extracted flexibility.
| Manuel Katholnigg manuel.katholnigg@tum.de | anytime | 2026-03 |
| MT, RI | Quantifying Data-Quality Thresholds for Energy Forecast Bootstrapping
Research method: Prototyping Research question: What are the minimum data-quality thresholds required to bootstrap reliable energy forecasts for newly installed PV or wind and storage systems lacking historical data? How can public datasets be automatically curated to generate accurate "Day 1" forecast models?
Possible approach: Develop an automated curation pipeline that uses GPS location and asset type to integrate public datasets (e.g., open climate data, TUM EMT open data collection). Train bootstrapped forecast models that rely entirely on this curated public data. Quantify the data-quality thresholds and establish a clean, publishable benchmark for energy forecast bootstrapping.
| Manuel Katholnigg manuel.katholnigg@tum.de | anytime | 2026-03 |
| BT, RI | Urban Wind Resource Characterization: Building a Real-Time Measurement Platform Research method: Experimental Research Questions: How significantly does the measured "urban" wind profile (turbulence intensity, gust factors) differ from standard weather station data or wind maps? Is this specific roof technically suitable for a wind turbine
Possible Approach: Plan and set up a wind measurement campaign and set up sensors Build a public open-data repository and live visualization dashboard - Establish correlation between modeled and measured wind
Evaluate the suitability of this location for a small wind turbine
| Jonas Betscher jonas.betscher@tum.de | anytime | 2026-02 |
| BT, RI | Benchmarking state-of-the-art deterministic Machine-Learning models for Wind Energy Forecasting Research method: Modeling Research questions: - What should a benchmarking dataset include?
- How do different models compare on forecast performance, reliability, complexity and computational costs?
Possible approach: - Research the state of the art in deterministic Machine-Learning models for wind energy forecasting
- Implement the models in a standardized way ready for benchmarking
- Create a benchmarking dataset
- Evaluate the models based on performance, reliability, complexity and computational costs
| Jonas Betscher jonas.betscher@tum.de | anytime | 2026-01 |
| MT, RI | Feature Engineering for Wind Energy Forecasting Research method: Modeling Research questions: - How do different factors influence wind power generation?
- How can feature engineering improve forecast performance, reliability, complexity and computational costs?
Possible approach: - Investigate the impact of key factors on forecasting performance
- Research and apply advanced feature engineering methods
- Employ state-of-the-art machine learning models to assess downstream impacts using metrics for predictive accuracy (e.g., MAE, MSE), reliability (e.g., calibration, sharpness), model complexity (e.g., parameter count), and computational costs (e.g., training time relative to baselines)
| Jonas Betscher jonas.betscher@tum.de | anytime | 2026-01 |
| MT, RI, Forschungspraxis | Using Deep Learning to Forecast Battery Aging Research method: Prototyping Research questions: - How can deep learning (in particular transformers) be used to forecast battery aging based on operation?
- How well do these methods perform compared to physics-based and existing ML-based benchmarks?
Possible approach: - Understand state-of-the-art based on literature
- Analyze existing open datasets to assess their use for ML experiments
- Evaluate existing models on selected datasets
- Use selection of datasets to fit novel ML models and compare the results
| Christoph Goebel christoph.goebel@tum.de | anytime | 2025-11 |
| MT, RI, Forschungspraxis | Using Flexibility of Production Systems for Energy Management Research method: Prototyping Research questions: - How can the flexibility of production systems (e.g., running machines slower or faster) be used to improve energy management (e.g., reducing cost by using more locally generated solar power or adjusting consumption to dynamic tariffs)
- How can corresponding optimization problems be defined and solved?
- How can different optimization methods be benchmarked?
Possible approach: - Understand state-of-the-art based on literature
- Define optimization scenarios (e.g., based on production use cases described in the literature)
- Develop efficient Python or Julia code to implement an optimization method
- Benchmark optimization method using baseline scenarios
| Christoph Goebel christoph.goebel@tum.de | anytime | 2025-09 |
MT, RI, Forschungspraxis | Extension of the Energy Management System Benchmarking Framework EMSx Research method: Prototyping Research questions: - How can EMSx be extended to include more sophisticated modeling capability (e.g., similar to OCHRE)?
- How can EMSx be extended to use other datasets and higher time resolution?
- How can EMSx be extended to enable benchmarking of reinforcement learning algorithms?
Possible approach: - Understand current EMSx framework and code (written in Julia)
- Develop efficient Julia code to implement selected extensions
- Evaluate extensions using actual data
| Christoph Goebel christoph.goebel@tum.de | anytime | 2025-03 |
MT, RI, Forschungspraxis | Solving Multi-Period Optimal Power Flow in Distribution GridsResearch method: Prototyping Research question: - Which methods can be used to calculate multi-period optimal power flow in distribution grids?
- How do these methods scale for different distribution grid sizes and scenarios?
Possible approach: - Understand state-of-the-art models based on literature
- Formulate mathematical optimization problem
- Develop efficient Julia code using state-of-the-art methods to solve the problem
- Evaluate solution method using realistic distribution grids with solar, load, and storage
| Christoph Goebel christoph.goebel@tum.de | anytime | 2025-03 |
MT, RI, Forschungspraxis | Global Forecasting Models for Low Voltage Load Forecasting Research method: Prototyping Research question: - How global forecasting methods be applied to load forecasting on the building level?
- How can their performance be evaluated?
Possible approach: - Understand state-of-the-art based on literature
- Develop efficient Python or Julia code to implement global forecasting models for electric load forecasting
- Evaluate method using actual data
| Christoph Goebel christoph.goebel@tum.de | anytime
| 2025-03 |
MT, RI, Forschungspraxis | Evaluation of MPC-based EMS on High Frequency Data Research method: Prototyping Research question: - How can MPC-based EMS designed to work on actual high-frequency data (1 sec - 1 min load and solar generation)?
- What is the trade-off between computational complexity and economic performance?
Possible approach: - Understand state-of-the-art based on literature
- Develop efficient Python or Julia code to implement MPC-based EMS that minimized cost of energy
- Evaluate method using actual data
| Christoph Goebel christoph.goebel@tum.de | anytime
| 2025-03 |
MT, RI, Forschungspraxis | Distribution Grid Model Generation Methods Research method: Prototyping Research question: - How can realistic distribution grid models be synthesized?
- How can synthesized distribution grid models be evaluated?
Possible approach: - Understand state-of-the-art based on literature
- Develop efficient Python or Julia code to implement distribution grid generation method
- Evaluate method using actual data
| Christoph Goebel christoph.goebel@tum.de | anytime | 2025-03 |
MT, RI, Forschungspraxis | Decentralized P2P Energy Trading Under Network ConstraintsResearch method: Prototyping Research question: - How can peer to peer energy trading in distribution grids be realized while respecting physical constraints?
- Which approaches exist?
- How can these approaches be benchmarked using realistic distribution grid models?
Possible approach: - Understand state-of-the-art based on literature
- Develop efficient Python or Julia code implementing existing methods
- Evaluate performance of methods using realistic models of distribution grids
| Christoph Goebel christoph.goebel@tum.de | anytime | 2025-03 |
RI | Data-Driven Optimization for a Real Agri-PV Site with Grid-Charged Battery Storage Research method: Prototyping Research question: Possible approach: Design a data-driven framework based on available streams of data such as energy production and performance metrics, weather and irradiance measured at the site Implement an optimization-based module that improves upon existing operational practices Benchmark performance against historical data of the site
| Arbel Yaniv arbel.yaniv@tum.de
| anytime | 2026-01 |
MT, RI | Generating Representative AC-OPF Datasets Research method: Prototyping Research question: - How to systematically generate representative AC OPF datasets that span the feasible region?
- How to measure the generated data level of representativeness?
Possible approach: - Understand state-of-the-art based on literature
- Enhance existing state-of-the-art methods to increase level of representativeness
- Benchmark against dataset generated with SOTA approaches
| Arbel Yaniv arbel.yaniv@tum.de
| anytime | 2025-08 |
MT, RI | Artificial Neural Networks for Power flow Research method: Prototyping Research question: - What model design is suitable for ANNs power flow predictions?
- What learning techniques could improve model performance?
Possible approach: - Understand state-of-the-art based on literature
- Implement ANN model for power-flow prediction
- Benchmark results against SOTA ANNs
| Arbel Yaniv arbel.yaniv@tum.de
| anytime | 2025-09 |
| BT | Spatiotemporal Assessment of Electric Vehicle Adoption Scenarios on the Distribution Grid Research method: Prototyping/ simulation Research question: - What are the impacts of different electric vehicle (EV) penetration levels on the reliable operation of distribution grids?
- How can various EV adoption scenarios be effectively modelled to capture realistic spatial and temporal load variations?
Possible approach: - Understand state-of-the-art based on literature
- Model realistic grid demand incorporating different EV adoption rates
- Analyse results to quantify implications on grid operation
Your background/interests: - Power-Flow analysis
- Programming experience - python
| | anytime | 2025-06 |
RI | Benchmarking Active-learning Approaches for Optimal Power-Flow Dataset Generation Research method: Prototyping Research question: - How to systematically generate representative AC OPF datasets that span the feasible region?
- What is the impact of the query approach on the model's predictive performance?
Possible approach: - Understand data sampling approaches based on literature
- Implement active learning pipeline (query strategy + ML model) to guide data sampling of OPF instances
- Benchmark various sampling approaches based on the scikit-activeml library
Your background/interests: - Python programming skills
- Machine learning
- Optimal power-flow
Resources: [1] Herde, Marek, et al. "scikit-activeml: A Comprehensive and User-friendly Active Learning Library." (2025). | Arbel Yaniv arbel.yaniv@tum.de | anytime | 2025-09 |
RI | Non-Intrusive Load Monitoring with Active Learning Research method: Prototyping Research question: - What is the impact of different acquisition functions on the performance of appliances disaggregation?
- What is the best fine-tuning approach, both layer-wise and sample-wise?
Possible approach: - Understand state-of-the-art based on literature
- Leverage scikit-activeml to analyse different active-learning approaches for NILM
- Try different fine-tuning experimental setups
Tanoni, Giulia, et al. "A weakly supervised active learning framework for non-intrusive load monitoring." Integrated Computer-Aided Engineering 32.1 (2025): 39-56. | Arbel Yaniv arbel.yaniv@tum.de | anytime | 2025-09 |
MT, RI | Designing domain-specific priors for Prior-Fitted-Networks in Energy Management Prior-Fitted-Networks [1] have shown intriguing success on small-scale tabular classification and regression tasks. Many energy management problems can be formulated as such classification or regression tasks. The TabPFN [2] uses a prior (=data generating mechanism) from either a Bayesian neural network or a structural causal model. These are designed for general tasks. The question is, can we achieve better performance by designing a prior that is more suited to a specific energy management task? Research method: Modeling/ prototyping Research question: - Can Prior-Fitted-Networks trained on domain-specific priors outperform TabPFN and/ or traditional ML models?
- Which priors are suitable for <your selected energy management task>?
Possible approach: - Understand prior-fitted-networks and TabPFN from the literature
- Select energy management task (thermal building modeling, solar forecasting, wind forecasting, load forecasting, NILM)
- Suggest & test different priors for PFNs
- Benchmark results against TabPFN and a traditional ML model
Your background/interests: - Interested in energy management and machine learning
- Statistics
- Programming experience - python
Resources: [1] Müller, Samuel, et al. "Transformers can do bayesian inference." arXiv preprint arXiv:2112.10510 (2021). https://arxiv.org/pdf/2112.10510 [2] Hollmann, Noah, et al. "Tabpfn: A transformer that solves small tabular classification problems in a second." arXiv preprint arXiv:2207.01848 (2022). https://arxiv.org/pdf/2207.01848 | Jan Marco Ruiz de Vargas janmarco.ruiz@tum.de | anytime | 2025-08 |
MT, RI | Improving ML models with pre-training from priors inspired by Prior-Fitted-Networks Prior-Fitted-Networks [1] have shown intriguing success on small-scale tabular classification and regression tasks. Many energy management problems can be formulated as such classification or regression tasks. Can the priors used in PFN training also be useful for improving traditional ML model performance on energy management tasks? Research method: Modeling Research question: - Does an ML model with access to the data from a PFN-prior achieve superior performance compared to one without?
- Which training strategies use this additional training data most optimally?
Possible approach: - Understand prior-fitted-networks and TabPFN from the literature
- Select energy management task (thermal building modeling, solar forecasting, wind forecasting, load forecasting, NILM) and ML model
- Suggest & test different pre-training strategies for the ML model
- Benchmark results against TabPFN and a traditional ML model
Your background/interests: - Interested in energy management and machine learning
- Statistics
- Programming experience - python
Resources: [1] Müller, Samuel, et al. "Transformers can do bayesian inference." arXiv preprint arXiv:2112.10510 (2021). https://arxiv.org/pdf/2112.10510 [2] Hollmann, Noah, et al. "Tabpfn: A transformer that solves small tabular classification problems in a second." arXiv preprint arXiv:2207.01848 (2022). https://arxiv.org/pdf/2207.01848 | Jan Marco Ruiz de Vargas janmarco.ruiz@tum.de | anytime | 2025-08 |
BT, IP | Applying Prior-Fitted-Networks to Energy ML Tasks Prior-Fitted-Networks [1] have shown intriguing success on small-scale tabular classification and regression tasks. Many energy management problems can be formulated as such classification or regression tasks. How good does a PFN then perform on energy ML tasks compared to other ML methods? Research method: Benchmarking Research question: - For <energy ML task>, how good does TabPFN or its variants perform on <energy ML task> compared to other state of the art methods?
- Possible energy ML tasks (select one!):
- Wind Power Forecasting (very short term = time series modeling)
- Wind Power Forecasting (numerical weather prediction → power = power curve modeling)
- Thermal building modeling
- Solar Power Forecasting
- Load Forecasting
- Non-intrusive load monitoring
Possible approach: - Understand prior-fitted-networks and TabPFN from the literature
- Select energy management task (thermal building modeling, solar forecasting, wind forecasting, load forecasting, NILM) and SOTA benchmark ML models
- Benchmark results
Your background/interests: - Interested in energy management and machine learning
- Statistics
- Programming experience - python
Resources: [1] Müller, Samuel, et al. "Transformers can do bayesian inference." arXiv preprint arXiv:2112.10510 (2021). https://arxiv.org/pdf/2112.10510 [2] Hollmann, Noah, et al. "Tabpfn: A transformer that solves small tabular classification problems in a second." arXiv preprint arXiv:2207.01848 (2022). https://arxiv.org/pdf/2207.01848 | Jan Marco Ruiz de Vargas janmarco.ruiz@tum.de | anytime | 2025-10 |
MT, RI, BT IP | Finetuning Multi-Modal LLMs on Energy ML Tasks Colleagues have fine-tuned MMLLMs on a tabular classification task [1], with significant improvement in prediction accuracy. This is a puzzling result, why does this work? Can it work with other prediction tasks? In this topic, we will investigate this by applying LLM finetuning to energy ML tasks. Research method: Benchmarking Research question: - For <energy ML task>, how good does a fine-tuned MMLLM perform on <energy ML task> compared to other state of the art methods?
- Possible energy ML tasks (select one!):
- Wind Power Forecasting (very short term = time series modeling)
- Wind Power Forecasting (numerical weather prediction → power = power curve modeling)
- Thermal building modeling
- Solar Power Forecasting
- Load Forecasting
- Non-intrusive load monitoring
Possible approach: - Understand how to finetune LLMs and how to properly feed it the relevant data.
- Select energy management task (thermal building modeling, solar forecasting, wind forecasting, load forecasting, NILM) and SOTA benchmark ML models
- Benchmark results
Your background/interests: - Interested in energy management and machine learning
- Statistics
- Programming experience - python
Resources: [1] Domiter, Andrea, and Srinivasan Keshav. "Machine Learning for Building-Level Heat Risk Mapping." Proceedings of the 16th ACM International Conference on Future and Sustainable Energy Systems. 2025. | Jan Marco Ruiz de Vargas janmarco.ruiz@tum.de | anytime | 2025-10 |
MT, RI | Adaptive Control of Electric Bus Depots for Grid Services
Keywords: Learning methods, electric buses, grid flexibility Objective:
Electric bus depots are considered as passive grid burdens. The objective of this MSc thesis will be to investigate how electric bus depots can be transformed into flexible assets deploying advanced control strategies for energy management.
Research questions:
- How real-time fleet management strategies improve the reliability of electric bus operations under practical system constraints?
- How do different real-time control approaches compare in managing disruptions in deadline-constrained fleet operations?
Expected background: - Interested in energy management and machine learning
- Mathematical modeling
- Programming experience (Python)
| | Anytime | 2026-01 |